NCP-GENL Practice Questions - NVIDIA-Certified Professional Generative AI LLMs Exam

It enforces a causal mask on future tokens It keeps the softmax in a regime with usable gradients as head dimension grows It prevents the model from learning positional information It reduces parameter count by sharing keys across heads